Panda Quality Score Audit
Patent Family: US9135307B2 — "Ranking search results" (Panda Algorithm, filed 2011) + US9031929B1 — "Site quality score" (Navneet Panda) Related Patents: US8707459B2 (content originality) · US8898296B2 (boilerplate detection) · US7734627B1 (duplicate detection) · US8150842B2 (author reputation) · US9558233B1 (dwell time) · WO2014209758A1 (above-the-fold) Source: Bill Slawski, SEO by the Sea
Philosophy: Think Like the Classifier
The Panda patent describes a decision tree trained on human quality rater judgments. It does not evaluate content subjectively — it extracts measurable features from pages and runs them through a classification model. Your audit must do the same.
Core things to understand:
- Panda is a pre-query classifier. Quality scores are computed before any search query is matched. The page either passes or it does not, regardless of what keyword it targets.
- Panda operates at both page level AND site level. A single bad page does not kill a site, but a high ratio of bad pages does.
- The patent uses ratios, not absolute numbers. A 50-page site with 45 quality pages is healthier than a 5,000-page site with 500 quality pages.
- The decision tree is sequential. Content uniqueness is evaluated first. If a page fails uniqueness, downstream signals matter less.
Core question the classifier asks:
"Would a human quality rater judge this page as providing genuine value that could not be trivially replicated by a template, scraper, or content spinner?"
The 8-Signal Decision Tree
Signal 1: Content Uniqueness
Patent: US8707459B2 — original-to-copied ratio
This is the first gate. Content uniqueness is evaluated before all other signals. A page that fails here gets classified as low quality regardless of everything else.
What to check:
- Unique sentence count vs. total sentences in the page
- Percentage of body text that appears elsewhere on the web (Copyscape, Siteliner, or manual search)
- Boilerplate percentage — how much of the page is navigation, footer, sidebar vs. main content
- If syndicating content: is the original source canonicalized back to your page?
- Paraphrased duplicates — content that mirrors another source at the word-swap level
Thresholds:
- 70%+ unique body content: PASS
- 50-70%: MARGINAL (monitor closely)
- Below 50%: HIGH RISK — content is predominantly copied or templated
Scoring: 0-20 points. 20 = fully unique original content. 10 = borderline. 0 = copied.
Signal 2: Ad-to-Content Ratio
Patent: WO2014209758A1 — above-the-fold content analysis
Google's system evaluates how much of the visible screen area (above the fold on standard viewport) is occupied by ads, promotions, or non-editorial content versus the actual content that satisfies the query.
What to check:
- Measure the first viewport (1024x768 standard): what percentage is ads?
- Count ad units in the entire page: how many per screen length?
- Are ads interspersed throughout content, interrupting reading flow?
- Is content immediately visible without scrolling, or do ads push it below the fold?
- Affiliate links density — are affiliate links the primary purpose of the page?
Thresholds:
- Below 25% ad space above fold: PASS
- 25-40%: MODERATE — acceptable but watch
- Above 40% above fold: HIGH RISK
Scoring: 0-15 points.
Signal 3: Expert Authorship
Patent: US8150842B2 — author reputation scoring
The Panda system incorporates author reputation as a quality signal. Pages with attributed, credentialed authors from known authoritative profiles score higher.
What to check:
- Is there a clear byline on the page?
- Does the author have a bio with verifiable credentials?
- Is the author's entity established in Google's Knowledge Graph?
- Does the author's documented expertise align with the content topic?
- Does the site have an author archive page with consistent publishing history?
- For YMYL content: are medical, legal, or financial credentials displayed?
Scoring: 0-15 points. 15 = credentialed author, KG entity, relevant expertise. 0 = no author attribution.
Signal 4: Referral-to-Page Ratio
This signal measures traffic quality relative to site size. The core insight: a site with 500 pages and 500 daily unique visitors is healthier than a site with 5,000 pages and 500 daily visitors.
What to check:
- Total indexed pages / unique visitors per day = referral-to-page ratio
- Identify pages that have never received organic traffic despite being indexed
- Calculate what percentage of your indexed pages receive at least one visit per month
- Pages with zero visits in 12+ months are quality score anchors
Formula: If more than 30% of indexed pages have received no traffic in 12 months, this is a high-risk signal.
Scoring: 0-10 points.
Signal 5: Site-Level Quality Score
Patent: US9031929B1 — the Navneet Panda site-level quality patent
The site-level quality score propagates across the domain. This is the most important signal for understanding Panda's site-wide behavior.
Formula from patent: (R0 + 1) / R1
- R0 = pages with positive referral signals (receive traffic, users engage)
- R1 = total crawled pages
What to check:
- Crawl all pages via Screaming Frog or similar
- Identify "anchor pages" — pages dragging down the site ratio
- Common anchor page types: thin category archives, auto-generated tag pages, low-traffic product variants, test pages, thin affiliate pages
- The fix: noindex anchor pages OR upgrade them to genuinely useful content
Scoring: 0-20 points (highest weight — site-level signals propagate to all pages).
Signal 6: Bounce Pad Detection
Source: Slawski, "How Google Might Filter Out Duplicate Pages from Bounce Pad Sites"
Bounce pad pages are designed to rank in Google and then immediately pass users through to another destination. They provide no value in themselves — they exist only as a traffic relay.
Red flags:
- Page content is thin (under 300 words) with a prominent call-to-action to leave
- Affiliate pages where all content leads to a merchant site
- "Review" pages where the only real content is an affiliate link
- Pages that automatically redirect after a short delay
- Pages where the primary CTA is "Visit Site" with no real informational content
Scoring: 0-10 points. If bounce pad pattern detected: 0.
Signal 7: User Engagement
Patent: US9558233B1 — dwell time / selection quality score
Google measures the time between when a user clicks a result and when they return to the SERP. Long dwell time = satisfied user. Immediate return = pogo-stick = dissatisfied user.
What to check:
- Average session duration for organic landing pages (Google Analytics)
- Bounce rate segmented by organic traffic source
- Scroll depth for key pages (does content pull users in or do they leave at the fold?)
- Pages-per-session for organic entrances
- Return visit rate from organic traffic
Pogo-stick pattern: If users consistently return to SERP within 10-30 seconds of clicking your result, this is a direct negative quality signal that suppresses rankings.
Scoring: 0-5 points.
Signal 8: Content Depth
Patent: US8898296B2 — boilerplate detection via DOM tree analysis
Google's system parses the DOM tree to separate boilerplate content (navigation, sidebar, footer, repeated elements) from main content. It then evaluates the depth and value of what remains.
What to check:
- Strip boilerplate from the page and read what's left — does it stand alone as valuable?
- Word count of main content only (not including nav/footer/sidebar)
- Does the content cover the topic with genuine depth, or does it skim the surface?
- Are there factual claims that could be verified and are they backed up?
- Does the content contain original analysis, examples, or data — or just restate common knowledge?
Thresholds:
- 1,000+ words of unique main content with genuine depth: STRONG
- 500-1,000 words with real value: ACCEPTABLE
- Under 500 words or thin coverage: THIN CONTENT FLAG
Scoring: 0-5 points.
Scoring Matrix
| Signal | Weight | Your Score |
|---|---|---|
| Content Uniqueness | 20 | /20 |
| Ad-to-Content Ratio | 15 | /15 |
| Expert Authorship | 15 | /15 |
| Referral-to-Page Ratio | 10 | /10 |
| Site-Level Quality Score | 20 | /20 |
| Bounce Pad Detection | 10 | /10 |
| User Engagement | 5 | /5 |
| Content Depth | 5 | /5 |
| TOTAL | 100 | /100 |
Score Interpretation
| Score | Verdict | Action |
|---|---|---|
| 80-100 | Quality Passes | Maintain and monitor |
| 60-79 | Moderate Risk | Address weakest signals within 30 days |
| 40-59 | High Risk | Site-level cleanup required immediately |
| Below 40 | Panda Suppression Likely | Significant site audit and content prune needed |
Site-Level Audit Process
When running site-level (not just page-level):
- Crawl the entire domain (Screaming Frog or Sitebulb)
- Segment pages into: High quality / Marginal / Thin / Anchor (zero-traffic)
- Calculate ratio: (high quality pages) / (total crawled pages) — target above 0.7
- Decision per anchor page: noindex it, improve it, or 301 redirect it to the nearest relevant quality page
- Never delete pages that have inbound links without redirecting
- Monitor in Search Console after changes — Panda recoveries typically take 3-6 months to show in rankings
Common Misdiagnoses
"My content is good, so Panda shouldn't affect me." Site-level quality score means your good pages share domain-level quality signals with your thin pages. You can have 50 excellent pages and 200 thin auto-generated pages — the thin pages drag down the whole domain.
"I'll just noindex everything." Noindexing thin pages is correct. But noindexing too aggressively removes pages Google might be crawling for link equity. The fix is surgical: noindex only zero-value pages, improve marginal pages.
"Panda was years ago." Panda signals are baked into Google's core ranking algorithm as of 2016. The patents remain active. The signals are evaluated continuously, not in discrete updates.